nWave-ai/nWave
AI agents that guide you from idea to working code, with you in control at every step.
nWave – AI‑driven feature‑delivery assistant for Claude Code
What it is
- A set of AI agents that live inside Anthropic’s Claude Code IDE and guide a developer through a structured, seven‑step “wave” workflow (discover → diverge → discuss → design → devops → distill → deliver).
- Each wave is handled by a dedicated agent that produces artefacts (docs, designs, tests, code) which the human reviews before the next wave starts. The system never runs end‑to‑end without human approval.
How it works
- Installation is a single script that adds the
nwave‑aiCLI and registers the agents as Claude Code plugins. The CLI can be installed withuv(preferred) orpipx. - Inside Claude Code you invoke agents with slash commands, e.g.
/nw‑discover,/nw‑design,/nw‑deliver. A concierge command/nw‑buddycan tell you which wave to start next. - A DES (Design‑Enforcement‑System) layer monitors every agent call, enforcing TDD phases, peer‑review requirements and other rigor profiles. It blocks unsafe edits and reports clear messages.
- Rigor can be tuned per project with
/nw‑rigor(lean, standard, thorough, exhaustive, or a custom profile), letting you balance token cost against quality.
Key concepts
| Wave | Typical entry point | Agent(s) | Goal |
|---|---|---|---|
| DISCOVER | Greenfield start | product‑discoverer |
Explore market/problem space |
| DIVERGE | Greenfield brainstorming | nw‑diverger |
Generate many possible solutions |
| DISCUSS | Any project | product‑owner |
Write requirements & user journeys |
| DESIGN | All | system‑designer, ddd‑architect, solution‑architect |
Architecture & domain model |
| DEVOPS | All | platform‑architect |
Infrastructure & deployment plan |
| DISTILL | All | acceptance‑designer |
Acceptance tests (Given‑When‑Then) |
| DELIVER | All | software‑crafter |
TDD implementation (RED→GREEN→COMMIT) |
Installation (5 min)
sh -c "$(curl -fsSL https://raw.githubusercontent.com/nWave-ai/nWave/main}/scripts/install/install.sh)"
- Requires Python 3.10+ and a running Claude Code instance.
- The script installs the CLI (
nwave‑ai) and copies the agents into~/.claude/. - Optional: use
uvfor faster, isolated installs;pipxworks too.
Typical first steps
- Open Claude Code and run
/nw‑buddy What should I do next?– the buddy tells you which wave to start. - Run the wave commands in order, e.g.:
/nw‑diverge "authentication approaches" /nw‑discuss "login with email & password" /nw‑design --architecture=hexagonal /nw‑distill "user‑login" /nw‑deliver - Review the artefacts each agent creates (markdown docs, design diagrams, test files) before approving the next wave.
Quality & token‑efficiency
- Rigor profiles let you pick a language model size (haiku, sonnet, opus) and decide whether to run mutation testing, peer review, etc.
- Lean mode uses the smallest model and skips reviewers, ideal for spikes; exhaustive mode uses the largest model and runs full mutation testing for production‑critical code.
Maintenance
- The CLI checks for updates when Claude Code starts; you can upgrade with
uv tool upgrade nwave‑ai && nwave‑ai install(or the equivalentpipxcommand). - Uninstall with
nwave‑ai uninstall– it removes agents, commands, and DES hooks but leaves your project files untouched.
Documentation & community
- Full docs at docs.nwave.ai – guides, reference, versioned per release.
- Quick‑start tutorial: Your First Feature.
- Discord community for support and discussion.
- Issues and contributions are handled via the GitHub repository.
Who it’s for
- Individual developers who want AI assistance but still retain control over every step.
- Teams looking to standardise a disciplined, AI‑augmented workflow (TDD, peer review, infrastructure as code) across projects.
- Organizations that need a way to enforce engineering best practices while leveraging Claude Code’s code‑generation capabilities.
All information above is taken directly from the repository’s README; no additional features have been inferred.
Related
- Project
- Project
- Project
- Project
- Project